چكيده انگليسي :
Landslides caused by slope failures result in significant casualties and damage to buildings and infrastructure. Earthquakes are a major triggering factor for landslides and slope failures. Therefore, accurately assessing the area affected by landslides is of paramount importance. The objective of this study is to classify the area affected by landslides based on slope characteristics and earthquake parameters. For this purpose, data on 90 earthquakes were collected. These data included two parameters related to slope characteristics: slope angle and the presence or absence of water. Five earthquake-related parameters were also considered: magnitude, intensity, epicentral distance, peak ground acceleration, and focal depth. Based on the statistical analysis of the collected data, the area affected by landslides was classified into three categories: less than 1000 km², 1000 to 10000 km², and more than 10000 km². Subsequently, five decision tree methods, including CART, J48, LMT, REP Tree, and Random Tree, were employed to develop models for classifying the affected area based on the aforementioned seven parameters. The models were evaluated using accuracy, precision, sensitivity, F-measure, Matthews correlation coefficient (MCC), and area under the curve (AUC) metrics. The results showed that J48 and LMT models achieved the highest accuracy and sensitivity, both at 92.3%. In terms of precision, F-measure, and MCC, the J48 model ranked first with values of 93.8%, 92.3%, and 89.1%, respectively, followed closely by the LMT model with values of 93.6%, 92.1%, and 88.9%. However, the LMT model had a significantly higher AUC value compared to the J48 model. The AUC values for LMT and J48 were 98.2% and 96.4%, respectively. On the other hand, the Random Tree model performed the worst in all evaluations. The accuracy, precision, sensitivity, F-measure, MCC, and AUC values for this model were 76.9%, 76.9%, 76.9%, 77.4%, 65.6%, and 85.4%, respectively. Overall, these results suggest that tree-based algorithms, particularly LMT and J48, are suitable options for classifying the area affected by landslides due to their simple structure, high interpretability, and the results obtained from different evaluation metrics.
